2.067 Beta: Behavior of enum and ref changed
Hi, following coding raises a compiler error with the beta of 2.067. Is this error intended or not? It is working if I change first line of main to: ulong bits; enum Bits: ulong { none = 0 } bool hasBit(ref ulong rBits, ulong rBit) { return cast(bool)(rBits & rBit); } void main() { Bits bits; hasBit(bits, Bits.none); } function app.hasBit (ref ulong rBits, ulong rBit) is not callable using argument types (Bits, Bits) Kind regards André
Re: 2.067 Beta: Behavior of enum and ref changed
On Tuesday, 10 March 2015 at 07:04:48 UTC, Andre wrote: Hi, following coding raises a compiler error with the beta of 2.067. Is this error intended or not? It is working if I change first line of main to: ulong bits; enum Bits: ulong { none = 0 } bool hasBit(ref ulong rBits, ulong rBit) { return cast(bool)(rBits & rBit); } void main() { Bits bits; hasBit(bits, Bits.none); } function app.hasBit (ref ulong rBits, ulong rBit) is not callable using argument types (Bits, Bits) Kind regards André It's because enums are not implicitly convertible to their base type. It was probably a compiler bug that it worked before. It's a regression however, so I'll file an issue in Bugzilla. In the meantime you can do: hasBit(cast(ulong)bits, Bits.none); Or just use a ulong as you mentioned.
Re: 2.067 Beta: Behavior of enum and ref changed
On Tuesday, March 10, 2015 08:19:27 Meta via Digitalmars-d-learn wrote: > On Tuesday, 10 March 2015 at 07:04:48 UTC, Andre wrote: > > Hi, > > > > following coding raises a compiler error with the beta of 2.067. > > Is this error intended or not? > > It is working if I change first line of main to: ulong bits; > > > > enum Bits: ulong > > { > > none = 0 > > } > > > > bool hasBit(ref ulong rBits, ulong rBit) > > { > > return cast(bool)(rBits & rBit); > > } > > > > void main() > > { > > Bits bits; > > hasBit(bits, Bits.none); > > } > > > > function app.hasBit (ref ulong rBits, ulong rBit) is not > > callable using argument types (Bits, Bits) > > > > Kind regards > > André > > It's because enums are not implicitly convertible to their base > type. It was probably a compiler bug that it worked before. It's > a regression however, so I'll file an issue in Bugzilla. In the > meantime you can do: > > hasBit(cast(ulong)bits, Bits.none); > > Or just use a ulong as you mentioned. enums _are_ implicitly convertible to their base type. e.g. this compiles just fine void main() { enum S : string { a = "hello", b = "world" } string s = S.a; } It's the base type that isn't implicitly convertible to the enum type. However, the code in question still shouldn't compile because while a Bits variable may be implicitly convertible to ulong, it _isn't_ a ulong, so passing it as a ref argument of type ulong isn't legal. Implicit conversions aren't used with ref. With ref, the type must match exactly. - Jonathan M Davis
Re: 2.067 Beta: Behavior of enum and ref changed
On Tuesday, 10 March 2015 at 08:37:46 UTC, Jonathan M Davis wrote: It's the base type that isn't implicitly convertible to the enum type. Err, yes. I had that the wrong way around. Anyway, I filed an issue. https://issues.dlang.org/show_bug.cgi?id=14269
Re: 2.067 Beta: Behavior of enum and ref changed
Thanks a lot! Kind regards André On Tuesday, 10 March 2015 at 09:25:02 UTC, Meta wrote: On Tuesday, 10 March 2015 at 08:37:46 UTC, Jonathan M Davis wrote: It's the base type that isn't implicitly convertible to the enum type. Err, yes. I had that the wrong way around. Anyway, I filed an issue. https://issues.dlang.org/show_bug.cgi?id=14269
Re: 2.067 Beta: Behavior of enum and ref changed
On 03/10/2015 01:37 AM, Jonathan M Davis via Digitalmars-d-learn wrote: > However, the code in question still shouldn't compile because while a Bits > variable may be implicitly convertible to ulong, it _isn't_ a ulong, In other words, the result of the implicit conversion is an rvalue, created on the spot, not the actual lvalue. References cannot be bound to rvalues. Ali
Re: 2.067 Beta: Behavior of enum and ref changed
On 03/10/2015 11:05 AM, Ali Çehreli wrote: In other words, the result of the implicit conversion is an rvalue Steven Schveighoffer says there is no rvalue in this case; "an enum is a derivative": https://issues.dlang.org/show_bug.cgi?id=14269#c14 Ali
Re: 2.067 Beta: Behavior of enum and ref changed
On Tuesday, March 10, 2015 13:26:00 Ali Çehreli via Digitalmars-d-learn wrote: > On 03/10/2015 11:05 AM, Ali Çehreli wrote: > > > In other words, the result of the implicit conversion is an rvalue > > Steven Schveighoffer says there is no rvalue in this case; "an enum is a > derivative": > >https://issues.dlang.org/show_bug.cgi?id=14269#c14 Implicit conversions result in rvalues, not lvalues. What's potentially different about enums is that underneath the hood, there's no difference between an enum and its base type - they're represented exactly the same in terms of bits; it's just the type system that treats them differently. So, the implicit conversion just changes how the type system treats it rather than the object's representation. So, it's possible to make it work so that a ref argument of the base type accepts a variable of the enum type, whereas that doesn't work with something like classes or even different integral types. An implicit conversion of an enum variable to its base type is one of the few cases where you even _could_ have an implicit conversion involving ref, but implicit conversions are supposed to result in rvalues as far as D's type system is concerned, so the fact that you could pass an enum value to a function taking its base type as a ref argument was definitely a bug. - Jonathan M Davis